Insights Into the Nature of the Transition Zone from Physically Constrained

Home , Mineral physics, Pyrolite

Insights into the nature of the transition zone from SPECIAL FEATURE physically constrained inversion of long-period seismic data

Fabio Cammarano* and Barbara Romanowicz

Berkeley Seismological Laboratory, University of California, 215 McCone Hall, Berkeley, CA 94720

Edited by Ho-kwang Mao, Carnegie Institution of Washington, Washington, DC, and approved February 7, 2007 (received for review January 4, 2007)

Imposing a thermal and compositional significance to the outcome of that a seismic model is not required to have a meaning in terms the inversion of seismic data facilitates their interpretation. Using of these physical parameters. long-period seismic waveforms and an inversion approach that in- Typically, the interpretation of a given seismic data set is cludes constraints from mineral physics, we find that lateral variations performed in two steps. First, a seismic model is constructed that of temperature can explain a large part of the data in the upper fits the data satisfactorily. Second, the model is interpreted in mantle. The additional compositional signature of cratons emerges in terms of temperature and composition based on the knowledge the global model as well. Above 300 km, we obtain seismic geotherms of the elastic and anelastic properties of mantle minerals plus that span the range of expected temperatures in various tectonic constraints on plausible composition and temperature ranges regions. Absolute velocities and gradients with depth are well con- from geochemistry [i.e., the signature of outcropping rocks strained by the seismic data throughout the upper mantle, except (orogenic peridotites) and mantle inclusions (xenoliths and near discontinuities. The seismic data are consistent with a slower xenocrysts)], heat flow measurements at the surface, and con- transition zone and an overall faster shallow upper mantle, which is ditions for melting mantle materials. The second part of the not compatible with a homogenous dry pyrolite composition. A process is commonly regarded as critical for interpretation. gradual enrichment with depth in a garnet-rich component helps to Together with the trade-off between temperature and compo- reduce the observed discrepancies. A hydrated transition zone would sition, other factors should be considered. Temperature- help to lower the velocities in the transition zone, but it does not dependent anelasticity implies nonlinearity of the partial deriv- explain the seismic structure above it. atives with temperature in the upper mantle (1). Consequently, absolute velocities are required for a correct interpretation. The mantle composition ͉ seismology ͉ mineral physics presence of mineralogical phase transitions further complicates the interpretation of the transition zone structure (14, 16). Finally, the large uncertainties that still characterize the elastic he thermal state and composition of Earth’s upper mantle and, even more, anelastic parameters derived by mineral physics and transition zone dictate its dynamics from microscale GEOPHYSICS T (see refs. 14, 17, and 18) hamper a precise interpretation. (e.g., creep mechanisms and earthquakes) to macroscale (e.g., Instead, it is often forgotten that the seismic models are already modality of mantle convection and plate tectonics). The knowl- an interpretation of the data. The seismic models are not unique, edge of these fundamental physical parameters and their 3D and they depend on the parametrization, distribution, quality, variations in Earth’s deep interior is indirect and relies entirely and type of data used. More importantly, a given best-fit seismic on the interpretation of geophysical data, based on insights from model does not necessarily correspond to a physical model. For theoretical and experimental mineral physics. Among these data, example, the previous work of Cammarano et al. (14) has pointed seismological observations constitute a main source of informa- out the difficulties of interpreting seismic reference models tion. Seismic waves record information about the elastic (and [Preliminary Reference Earth Model (PREM) (19) and AK135 anelastic) structure of Earth. Long-period seismic data provide (20)] in terms of temperature and composition. The same the most comprehensive global constraints on upper mantle problems are propagated to 3D tomography models that are shear velocity structure. Fundamental-mode surface waves are obtained by perturbing a starting average model. mostly sensitive to the uppermost mantle structure, and includ- To overcome this problem, it is indeed important to test a ing overtones provides resolution in the transition zone. physical hypothesis directly against the seismic data (as in refs. 21 Lateral temperature variations in the Earth have been inferred and 22). The general idea is to use the mineral physics information since the first tomographic studies began to reconstruct 3D already at an early stage of the seismic data processing, hence seismic velocity structure (1, 2). However, despite the ever- imposing a physical significance to the seismic tomography model. improving resolution of seismic velocity models and a general Here we invert long-period seismic waveforms, which are mainly agreement of different models on at least the large-scale struc- sensitive to shear velocity, with respect to a physical reference ture (e.g., refs. 3–6), interpretation is still challenging. The few model instead of a standard 1D seismic reference model (e.g., quantitative interpretation efforts made to date have shown that PREM). Velocity variations will thus correspond to thermal (or much of the seismic heterogeneity in the uppermost mantle is compositional) variations and a consistent 3D density and VP probably thermal in origin (7, 8), but it is likely that there is a compositional contribution under cratons (9, 10) and that fluids influence the very low velocities in the mantle wedge above Author contributions: F.C. and B.R. designed research; F.C. and B.R. performed research; subduction zones (8). By adding constraints from gravity or F.C. and B.R. analyzed data; and F.C. and B.R. wrote the paper. geoid data, joint inversions for seismic velocity and density have The authors declare no conflict of interest. been performed because of the better chance to isolate thermal This article is a PNAS Direct Submission. and compositional effects (11–13). Whereas temperature sensi- Abbreviations: PREM, Preliminary Reference Earth Model; PREF, Physical Reference. tivity dominates for seismic velocities, especially at shallow *To whom correspondence should be addressed. E-mail: [email protected]. depths (7, 14, 15), density also strongly varies with composition. This article contains supporting information online at www.pnas.org/cgi/content/full/ Besides the trade-off between temperature and composition, 0608075104/DC1. an important issue for seismic interpretation concerns the fact © 2007 by The National Academy of Sciences of the USA

www.pnas.org͞cgi͞doi͞10.1073͞pnas.0608075104 PNAS ͉ May 29, 2007 ͉ vol. 104 ͉ no. 22 ͉ 9139–9144 Downloaded by guest on September 27, 2021 structure may be determined. We start by assuming purely thermal Events with moment magnitude larger than six and stations at variations because, in the upper mantle, composition plays a epicentral distance between 15° and 165° were selected. The secondary role. We compare the thermal features constrained by original seismograms have been deconvolved for the instrument the seismic data against expected thermal variations and estimated response and filtered between 60 s and a variable maximum geotherms in various tectonic regions. The average velocity model period, typically between 220 s and 1 h, which was chosen extracted from the physically constrained tomography provides according to the event magnitude. Wave-packets for both fun- insights into the nature of the transition zone. To assess the damental and overtones were then extracted from the seismo- uncertainties in the thermal model and estimate how well average grams based on the comparison of the observed trace with the velocity and velocity gradient with depth are constrained by long- PREM synthetics (5). The PREM represents very well the period seismic data in the upper mantle, we perform a series of average seismic structure of Earth. The bounds of the selection inversions starting with various models. criteria are large enough to make the procedure appropriate also for the alternative average upper mantle structure based on a Seismic Procedure physical model. Minor and major arc paths were selected for our The Starting Physical Model. Because seismic inversion is based on inversion. In total, the data set amounts to 39,829 fundamental perturbation theory, to be used as a reference model, the starting waveforms and 59,831 overtones. The windowing scheme not physical model must have an overall good fit to seismic data, only reduced the computation time of the inversion, but it comparable to the fit of radially symmetric seismic models (e.g., allowed the application of appropriate weights to different PREM). We tested our inversion by using some of the Physical phases. In our case, we assigned a double weight to the overtones Reference (PREF) models from Cammarano et al. (22). These are compared with the dominant fundamental modes. Noise and isotropic models with the same thermal and compositional struc- path redundancy were also considered in the weighting scheme ture, respectively 60-million-year-old oceanic geotherm in the litho- as described by Li and Romanowicz (6). sphere and a 1,300°C adiabat below that and pyrolite composition. The seismic waveform tomographic method is based on the The set of elastic and anelastic parameters for the mantle minerals Nonlinear Asymptotic Coupling Theory method (25), which is a were obtained from a Monte Carlo search within estimated bounds normal-mode perturbation approach that accounts for coupling from ref. 14. These bounds encompass the total range of the both along and across dispersion branches. The 2D broadband experimental values existing at the time of the compilation (3). sensitivity kernels computed with this methodology reproduce Only the models that fit seismic travel times (P and S arrivals) and the body-wave character of the seismic wave propagation and are fundamental mode eigenfrequencies as well as the seismic refer- therefore able to accurately model the overtones. ence models have been selected. Despite a variety of possible combinations of mineral physics parameters found (see ref. 21), the Inversion. We started the inversion with a low-resolution version PREF models are very similar seismically [supporting information of a recent anisotropic model (SAW642AN; ref. 5), but we (SI) Fig. 4], and they are very different from the seismic reference replaced the background seismic reference model with a physical models. reference model (i.e., PREF). SAW642AN is parametrized in For example, all of the models have a low velocity zone in the spherical splines with 642 equally spaced horizontal nodes and 16 uppermost mantle, where temperature effects prevail over pressure vertical nodes from the top of the mantle to the core–mantle effects. The models do not have any sharp discontinuity at 220 km boundary. We reparametrized the model to 162 horizontal as in PREM; they have a larger jump than seismic reference models nodes, which corresponds to a spatial resolution of Ϸ3,500 km, Ϸ410 km, associated with the olivine–wadsleyite phase transition, and we kept the original vertical parametrization. At this reso- and a faster transition zone. All these features are governed by the lution (corresponding to ϳ12° in a spherical harmonics param- given composition and thermal structure plus the strong constraint etrization), global models from different groups are very similar. on the integrated average of upper mantle velocities given by the We inverted for the isotropic part of the model down to 1,000 km, teleseimic travel times. A slower transition zone, balanced by an and we kept fixed the radially anisotropic part of the model, ␰ 2 2 overall faster upper mantle above, would explain the travel time represented by (VSH/VSV). Radial anisotropy is required to data better, but the mineral physics constraints do not allow that (SI simultaneously fit spheroidal and toroidal fundamental modes. Fig. 4) (22). We corrected for crustal structure by assuming the model Because of the sparse coverage of travel times at far-regional CRUST2.0 (26). The inversion procedure consists of several distances (epicentral distances Ͻ30°), it was not possible to iterations. For each step, we solved the inverse problem by using determine in that study (22) if such transition zone structure was the classical least-squares approach (27). Regularization required at a global scale. With the long-period overtone data schemes involve scaling the equations according to the weighting used here, we were able to better constrain the average structure scheme discussed and then damping. The number of iterations of the transition zone and draw conclusions on its nature. required to reach a variance reduction similar to the starting We selected three PREF models that represent well the model SAW642AN range from four to 11 iterations for the differences in seismic and mineral physics properties within the different PREF models tested. The selected models do not fit family of PREF models. The use of different starting PREF both spheroidal and toroidal modes equally well. Because we do models, characterized by their own sensitivities of seismic ve- not vary the radial anisotropic structure, we map this discrepancy locities to temperature, helped assess the effects of mineral into the isotropic part and therefore into temperature. More physics uncertainties on the physically constrained inversion. discussion about the trade-offs with anisotropy structure follows At the same time, this method tested the assumption that our (see Absolute temperatures below). outcome does not depend on the starting model. To further Although we did not invert for seismic attenuation, it is investigate this point, we also inverted the seismic data starting with important to account for the effects of anelasticity on seismic a model that has a much smaller jump at 410 km and different velocities to retrieve a correct thermal interpretation. Pressure- velocity gradients above and below this discontinuity (see Fig. 3, and temperature-dependent anelasticity, defined as for the model B). PREF models (14, 21) is included in our thermal models.

Data and Method. The data we used included three-component, Results long-period fundamental waveforms and higher-order mode Thermal Model. Lateral variations. The lateral variations in temper- surface waveforms collected for global tomography efforts over ature (Fig. 1) confirm the primary role of temperature in deter- the years at the University of California, Berkeley (5, 6, 23, 24). mining the seismic structure of the upper mantle and transition

9140 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.0608075104 Cammarano and Romanowicz Downloaded by guest on September 27, 2021 100 km ( -945.41/347.29 ) 300 km ( -317.91/252.87 ) a SPECIAL FEATURE

150 km ( -1034.78/318.11 ) 350 km ( -333.22/271.12 )

b 0

200 km ( -584.04/253.01 ) 450 km ( -224.27/174.99 ) 50 90 40 100

150

200 250 km ( -321.80/235.12 ) 550 km ( -213.72/130.62 ) 60 My old ocean 250 12 My old ocean Depth (km) 100 My old ocean 300 90 mW/m2 heatflow 40 mW/m2 heatflow 350 400 600 800 1000 1200 1400 1600 T (oC)

-1000 -300 0 300 Fig. 2. Seismic geotherms averaged over cratonic and oceanic regions. ∆Τ (Κ) Selected cratonic regions are represented by solid areas (a), and respective geotherms are in solid lines (b). Oceanic regions are represented with dashed Fig. 1. Lateral temperature variations inferred from the physically con- lines (a) and solid lines (b). strained waveform inversion. Positive anomalies in the uppermost mantle correspond to oceanic ridges, and negative anomalies correspond to cratonic areas. Negative thermal anomalies related to subduction zones appear below. steady-state and are based on surface heat flow and radiogenic heat The minimum and maximum values of anomaly compared with the mean production in the crust (31). Cratonic geotherms span a larger temperature value is given for each depth. range than the oceanic ones, probably mostly due to the compositional differences between them, suggesting that it is necessary to GEOPHYSICS zone. The thermal variations that explain the seismic data are include compositional variations in these regions. We sometimes indeed consistent with the range of temperatures expected from found extremely low temperatures in cratons (Fig. 2) that are likely geodynamics modeling (28). However, the secondary composi- due to the secondary compositional component as well. Also in this tional effect clearly emerges when one looks at the very low (i.e., as case, the mineral physics uncertainties do not significantly affect the low as 460 K at 100 km and 500 K at 150 km) values of temperature interpretation. Variations in the inferred geotherms obtained from obtained in the first 250 km beneath cratons. If more depleted different PREF starting models are generally small, and gradients composition (e.g., harzburgite) is considered for those regions, we with depth are similar. We found temperature variation as large as expect a shift upward of the lowest temperatures of Ϸ200° (14, 29). Ϸ75°C. Unlike in velocity models, the average temperature at a given depth The capacity of inferring absolute temperatures and thermal is not balanced around the mean value but is shifted toward the hot gradients, however, depends on how precisely we can determine regions (Fig. 1) because of the temperature-dependent anelasticity, absolute seismic velocities and gradients. To this end, we should which is included in our models. Accounting for its effects on point out that our results rely on assumed crustal and anisotropy seismic velocities helps to retrieve a more realistic thermal inter- structure. Changes in the crust may affect the uppermost mantle. In pretation. In particular, because of different thermal structures addition, accounting for lateral variations in the crustal kernels may beneath oceans and continents, attenuation is expected to have a also modify the upper mantle structure down to 300 km (32). Even very strong lateral heterogeneity, which is consistent with seismic more important are the trade-offs between the isotropic and attenuation global models (24). In our derived attenuation model, anisotropic parts of the model. The assessment of the trade-offs is the lateral variations depend only on temperature; therefore, they not a simple task and is not addressed here thoroughly (see ref. 33). resemble exactly the thermal variations shown in Fig. 1. In general, because of the sensitivity of the data used, the trade-offs The lateral thermal variations obtained by starting the inver- with anisotropy hamper a precise estimation of absolute isotropic sion with the other two PREF models are very similar, indicating velocities in the upper 300 km of the mantle. As a consequence, the that the additional uncertainties in the temperature partial thermal interpretation may change locally if a different anisotropic derivatives do not significantly affect the interpretation. Specif- structure is assumed. For example, variations as large as hundreds ically, these uncertainties affect it much less than regularization of degrees and small (but dynamically important) changes in schemes required by the inversion process. thermal gradients may result for the seismic geotherms shown in Absolute temperatures. The upper mantle geotherms constrained by Fig. 2 when using another anisotropic model (we tested a previous long-period seismic data are able to reproduce the range of version of the SAW642AN model; unpublished data). expected geotherms beneath oceans and cratons (Fig. 2). We In the transition zone, anisotropic effects have much less compared the seismic geotherms with geotherms for different influence on the results. A purely thermal interpretation with oceanic ages, which are based on the plate cooling model (30), and our pyrolitic models requires unrealistic thermal profiles. This is purely conductive continental geotherms, which were computed at discussed in a separate section.

Cammarano and Romanowicz PNAS ͉ May 29, 2007 ͉ vol. 104 ͉ no. 22 ͉ 9141 Downloaded by guest on September 27, 2021 0 tinuity are not well constrained by those data, which can be clearly PREM seen when we compare the gradients near the 410-km discontinuity 100 STARTING MODELS between the PREF inverted model and model B with the small PREF jump at 410 km (see Fig. 3). 200 model B Compared with the starting PREF model, the inverted model INVERTED MODELS is characterized by a faster upper mantle, a slower transition 300 PREF model B zone, and different depth gradients throughout the upper man- 400 tle. We found high absolute shear velocities at Ϸ300 km. Around this depth, the velocity gradient is also much higher than the Depth (km) 500 starting model (Fig. 3). The VS gradient between 500 and 600 km is slightly lower than PREM and more similar to the starting 600 velocity gradient of the adiabatic pyrolite model (Fig. 3). Overall, these features help to improve the fit to mode data for 700 4.2 4.4 4.6 4.8 5.0 5.2 5.4 5.6 5.8 6.0 both toroidal and spheroidal branches (see SI Fig. 6). The PREF VS (km/s) model(s) fit the fundamental modes satisfactorily, but the overtones, which are more sensitive to transition zone structure, are Fig. 3. Inverted average structures for a PREF model (red) and a model with poorly fit (SI Fig. 6). This discrepancy supports the results of different gradient and smaller jump at 410 km (blue; model B in Table 1). Solid Cammarano et al. (22), who suggested the need for a slower lines are the starting models; dashed lines are the inverted models. Circles on transition zone for improving the travel-time data fit and extended dashed lines indicate depths for which averages from the 3D model have been it to the global scale. Note that both the starting PREF and the computed. Velocity models beneath the oceanic (thin dashed lines) and cratonic ␰ (thin solid lines) regions of Fig. 2a are shown in the background in SI Fig. 5. inverted model are isotropic. If the radial anisotropy parameters and ␩ (as defined in PREM) (19) are added assuming PREM values, the overall fit further improves (SI Fig. 6). The Average Model. In Fig. 3, we show the inverted average 1D VS structure obtained from the 3D tomographic model. The average The Nature of the Transition Zone. The seismic features obtained for an extremely different starting model also is shown. The other from our physically constrained inversion are different from the PREF models are similar to the one presented; hence, they have ones in the common seismic reference models. Absolute veloc- similar inverted structure (VS averages for different depth ranges ities and gradients with depth are well constrained throughout are shown in Table 1). For a given anisotropy and crustal structure, the upper mantle, except near discontinuities. Despite the large absolute velocities and gradients are well constrained from long- uncertainties of the seismic properties of mantle minerals (see period data, except in the vicinity of the 410 km discontinuity ref. 22), our observations are robust enough to draw important (compare the two inverted models in Fig. 3; see also Table 1). conclusions on the nature of the upper mantle. Trade-offs with anisotropy that are important for the upper mantle The velocity gradient with depth between 250 and 350 km (see do not significantly affect the velocities and gradients below 300 km. Fig. 3) cannot be explained with any realistic thermal structure, We found that even a strongly anisotropic radial model, with assuming than composition does not change with depth. The variation from isotropy from ϩ2% to Ϫ2% along the transition velocity gradients of the PREF models within this depth-range zone, does not greatly change the isotropic features of the inverted are always similar (see SI Fig. 4), mainly because a relatively models. The regional velocity models beneath the same oceanic and narrow range was chosen for the shear pressure derivatives of cratonic regions as in Fig. 2a are shown in thin lines in SI Fig. 5. The olivine (GЈ was allowed to vary between 1.3 and 1.5; see table 1 velocity structure at Ͻ300 km is very similar in these different of ref. 14). To reproduce this gradient with a 1,300°C adiabatic tectonic regions, indicating the global validity of the inverted pyrolite, however, the values required of GЈ are too high. If we average model in the transition zone. vary only the olivine-GЈ (all other values are from table 1 of ref. The inverted model tends toward PREM, which is confirmed to 14), a value of 2.5 is required. Thermal gradients are much less be a very good average seismic model, but there are striking effective in reproducing this gradient. If we assume an isotherm differences due to the different starting parametrization. For (e.g., 1,700 K) instead of an adiabat, we still obtain a very high example, because PREF does not have a 220-km discontinuity, as GЈ (Ϸ2.35). Although further experiments on the shear prop- PREM does, the final model does not have this feature either. erties are warranted, such high values are very far from the Instead, a velocity gradient higher than the starting PREF model experimental values available until now (e.g., refs. 34 and 35). is required by the seismic data around this depth (Fig. 3). In Another important aspect of the seismic features is the charac- addition, long-period seismic waveforms are not directly sensitive to teristic of the seismic jump at the 410-km discontinuity. The velocity the mantle discontinuities. Therefore, the gradients near a discon- jump imposed in the starting model at the olivine–wadsleyite

Table 1. Upper mantle VS averages (in m/s) Depth range, km PREF PREF-b PREF-c Model B PREM AK135

Before inversion 75–150 4,407.4 4,419.5 4,433.7 4,413.4 4,456.8 4,508.2 200–350 4,562.6 4,559.9 4,566.1 4,635.7 4,650.7 4,636.9 450–600 5,387.8 5,380.6 5,390.1 5,309.6 5,297.2 5,323.8 After inversion 75–150 4,462.7 4,470.0 4,484.1 4,465.9 — — 200–350 4,615.6 4,611.1 4,584.1 4,600.8 — — 450–600 5,263.9 5,265.5 5,267.6 5,274.1 — —

Note that AK135 does not have the 220-km discontinuity as PREM does, and it has continental structure for the ﬁrst 150 km (20).

9142 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.0608075104 Cammarano and Romanowicz Downloaded by guest on September 27, 2021 transition dictates the gradients around it (Fig. 3). Pyrolitic models, proposed in the frame of a mantle evolution that leads to a chemical which are Ϸ60 vol % olivine, have a larger jump than seismic differentiation of the shallow upper mantle from the transition zone SPECIAL FEATURE reference models for this discontinuity [⌬VP, 6–10%; ⌬VS, 6–12% (43). A chemical layering of the mantle is supported by several (21); vs. 2.5% and 3.5%, respectively, for PREM and 3.5% and 4% geochemical arguments. For example, studies of basalts worldwide for AK135]. If we assume in our starting model the large jump indicate the need for different sources that have remained separate inferred from mineral physics, long-period seismic waveforms do over a long time (Ͼ1 billion years) (45). The presence of such a require the changes in gradients just above and below the 410-km chemical boundary implies layered mantle convection or an inef- discontinuity shown in Fig. 3 (inverted PREF). A thermal inter- ficient mixing mechanism. But seismic tomography provides clear pretation of this structure is not feasible. By using our preferred evidence for subduction reaching far into the lower mantle (4), and pyrolitic models, we obtain temperatures of Ϸ1,250 K at 350 km and modeling indicates that whole-mantle convection is sufficiently Ϸ1,750 K at Ϸ500 km and geodynamically unrealistic thermal vigorous to mix large-scale heterogeneities (2). Several suggestions gradients throughout the upper mantle. have been made to explain the discrepancy between geophysical Finally, the velocities in the transition zone are lower than for and geochemical evidence [see review by van Keken et al. (46)]. our preferred pyrolitic models. Note, however, that uncertainties Although the nature of mantle heterogeneities is still debated, there in mineral physics are large for seismic properties of the tran- is a general consensus on excluding the 410-km discontinuity as a sition zone (see Discussion). possible chemical boundary layer [with the exception of the water- We conclude that dry pyrolite cannot be reconciled with filter hypothesis (47), which we discuss below in this section]. seismic data. This conclusion has a global character, as the Furthermore, if the whole upper mantle, and not only the transition similarity of the velocity profiles at Ͻ300 km beneath oceans and zone, is assumed to be piclogitic, the seismic velocities would be cratons show (SI Fig. 5). systematically lower than pyrolite. To reach similar average VS as pyrolite, the temperatures should be lower overall. Assuming Discussion adiabaticity, we estimate a 150 K cooler potential temperature for Uncertainties in Mineral Physics Data. Although progress has been the upper mantle. This estimate is in conflict with the requirement made in experimental mineral physics and theoretical computa- of having a higher degree of partial melting than pyrolite to form tions of elastic properties at high pressure and temperature, there basalts (48). are still many uncertainties. In particular, shear properties of upper Small- and moderate-scale upper mantle compositional het- mantle minerals are difficult to measure at appropriate pressure erogeneities (102 to 105 m), which are beyond the resolution of and temperature ranges. For example, bulk and shear temperature- the seismic data used here, have been inferred from geochemical derivatives of the magnesium end-member of wadsleyite are still studies (see ref. 49 and references therein). It has been suggested debated (ѨKS/ѨT and ѨG/ѨT are Ϫ0.016 and Ϫ0.012 GPa/K, re- that the continuous reintroduction of heterogeneity in the upper spectively, in ref. 37 and Ϫ0.012 and Ϫ0.017 GPa/K, respectively, in mantle by subduction and partial melting may determine a ref. 38). The result is that absolute seismic velocities of mineralog- nonequilibrium state of the mantle, the nature of which could be ical models for a given composition and thermal structure are still more complex than what is possible to infer from available very different (see refs. 14, 16, 39, and 40). However, velocity modeling of mantle mixing (49). A mixture of enriched and gradients with depth along a mantle adiabat are very similar in depleted lithologies (e.g., eclogite plus harzburgite) instead of a GEOPHYSICS pyrolitic models because of the similar behavior of mantle minerals unique mineral assemblage (e.g., pyrolite) may occur in the when they are compressed adiabatically. A third-order Birch– upper mantle. Such a hypothesis, however, does not accommo- Murnagham equation of state with similar pressure derivatives (4–5 date the discrepancies we observe if the average chemical for KS and 1.5–2.5 for G) is usually appropriate to fit mineral physics composition of the upper mantle is pyrolite. It has been proposed data within the upper mantle pressure range (see, for instance, that small amounts of garnet pyroxenite in the upper mantle compilations in refs. 14, 16, and 39–41). The elastic properties of (Ϸ5%) can explain the garnet signature in middle ocean ridge olivine and wadsleyite at high pressure have been studied exten- basalt (50). This percentage is a lower limit, because the melt sively and are known well enough to constrain the jump at the phase derived from a peridotitic source with such proportions of transition (see ref. 35). Note that a larger jump than PREM at the pyroxenite will be largely dominated by the pyroxenite. A olivine–wadsleyite transition is consistently found in all of the gradual enrichment in the olivine-free (eclogite or pyroxenite) available pyrolitic models. Considering all of the uncertainties of component in the transition zone and right above it would help the elastic data, even the large uncertainties surrounding the shear to reduce the discrepancies we observe. A garnet-rich compo- pressure and temperature derivatives, the ⌬VS jump at the phase sition would help increase seismic velocities and reduce the jump transition has been found not Ͻ6% for pyrolite along a 1,300°C at the olivine–wadsleyite transition. For example, shear veloci- mantle adiabat (21). [Note that PREF models have a ⌬VS jump ties for pyroxenite composition reach values of Ϸ4.8 km/s at a closer to the average, Ϸ8.5% (see SI Fig. 4).] We found that a model depth of 350 km (51), which is consistent with the absolute VS with the same average VS and a ⌬VS jump of 6%, with related slower values required by long-period data (see Fig. 3). At this time, it transition zone and faster structure above it, is still not compatible is not clear whether such compositional gradation is dynamically with a purely thermal interpretation of the seismic data. The best possible. Note, for example, that the middle ocean ridge basalt way to further reduce the jump, as indicated by Duffy et al. (35), (i.e., basalt) composition below Ϸ280 km is up to 5% denser than should be to have a much higher shear temperature-derivative of an olivine-rich lithology (Ͼ40 vol %), although it becomes wadsleyite relative to olivine. Further studies are required to lighter than that composition at Ϸ660 km (3). Recent thermo- investigate this point, although the measured values on a large range chemical models (52), however, obtained such a compositional of silicates (see ref. 35 and the available data on wadsleyite, e.g., refs. gradient. In terms of average composition, the hypothesis of a 8 and 38) suggest that such a case is very unlikely. garnet-rich component in an olivine-rich matrix is similar to a piclogitic transition zone, but it does not prescribe the existence If Not Dry Pyrolite, Then What? A piclogite composition, character- of a chemical boundary at 410 km, and it is more flexible for ized by more garnet and pyroxenes and less olivine, has been producing the observed geochemical heterogeneity (49). proposed to provide a better fit to seismic 1D models in the Recently, it has been found that a large amount of water can be transition zone (6). The original piclogite was only 16 vol % olivine trapped in the nominally anhydrous minerals of the upper mantle. (42). Based on later estimates of garnet elasticity at high pressure, In particular, wadsleyite and ringwoodite can contain up to 3 wt % a more olivine-rich mineral assemblage has been suggested (Ϸ30% water (53) at pressures corresponding to transition-zone depths. in ref. 43 and Ϸ40% in ref. 44). The piclogite model has been This capacity tends to decrease as temperature increases, but it has

Cammarano and Romanowicz PNAS ͉ May 29, 2007 ͉ vol. 104 ͉ no. 22 ͉ 9143 Downloaded by guest on September 27, 2021 been estimated that Ϸ1 wt % can be accommodated into the but the high velocities and gradients above the wadsleyite stability wadsleyite and ringwoodite structure at Ϸ1,400°C (54). Based on field are not consistent with this hypothesis. stability and melting relations of mantle phases (55), a hydrous The interdisciplinary approach used here is general, and it can be transition zone that acts as a filter has been hypothesized by applied to any seismic data set. Besides facilitating the interpreta- Bercovici and Karato (47) to reconcile geochemical and geophysi- tion of seismic data, it permits us to relate several parameters (that cal observations. The elastic properties of hydrous minerals are are not directly constrained by the seismic data used) to the same expected to change with the concentration of hydrogen (56), and, physical structure. For instance, the physical model is characterized indeed, some of the recent experimental studies showed this (see, by a given set of partial derivatives with respect to temperature (and for example, refs. 36 and 57). Seismic velocities may be further composition) for seismic velocities and density. Even if the seismic reduced because of the dissipative effects of enhanced anelasticity data used are mostly sensitive only to shear velocity, the inverted 3D (3). We note, however, that we found high velocities and gradients VS model can be directly related to 3D thermal structure (neglecting in the upper mantle above the 410-km discontinuity. Even consid- as a first guess the compositional signature), but also VP-, density-, ering the large uncertainties of elastic data, it is difficult to explain and temperature-dependent attenuation structures can be pre- these features without invoking a compositional variation also dicted. By using this method, the same physical hypothesis can be above 410 km. tested against independent geophysical data. For example, dynamic topography or geoid anomalies can be used in principle to adjust the Conclusions physical structure based on seismic data. Other seismic data that complement the resolution of long- We show the feasibility of inverting seismic waveform data directly period waveform data can be included in the inversion. In for physical parameters (temperature and composition) in the particular, seismic data that are sensitive to mantle discontinui- upper mantle. Lateral variations in temperature explain most of the ties, such as PP and SS precursors, and short-period reflected, data. Secondary compositional features, associated with cratons, refracted, and converted phases (as in ref. 58) would provide a emerge from the inversion as well. Absolute velocities in the shallow better characterization of the seismic structure around the upper mantle depend on the crustal and anisotropic structure discontinuities. More investigation on shear properties at ap- adopted. Hence, a detailed interpretation in terms of absolute propriate pressure and temperature conditions for both anhy- temperatures and their gradients with depth rely on these features drous and hydrous phases of the transition zone minerals would as well. The seismic geotherms we obtained are consistent, how- help to confirm or refine our conclusions. ever, with expected geotherms in various tectonic settings. Includ- In conclusion, we argue that a truly interdisciplinary approach ing temperature-dependent anelasticity into the physical model is a that merges seismic observations, mineral physics data, and key point for correctly interpreting the seismic data. geodynamics constraints provides a valuable tool for obtaining Below 300 km, average velocities and gradients with depth are a consistent physical model of the Earth’s mantle. well constrained by the long period data used, except near the mantle discontinuities. We found that seismic data globally require We thank Federica Marone, Saskia Goes, Lars Stixrude, and Carolina a slower transition zone and an overall faster shallow upper mantle Lithgow-Bertelloni for fruitful discussions; Federica Marone for con- than reference pyrolitic models. Despite the large uncertainties in stantly helping with the seismic codes; Lars Stixrude and Carolina mineral physics data, the observations are not compatible with a Lithgow-Bertelloni for providing their mechanical–mixture model for pyrolite composition; and two anonymous reviewers for comments that thermal interpretation assuming dry pyrolite. A gradual enrich- improved the manuscript. This material is based on work partially ment in a garnet-rich component throughout the upper mantle may supported by National Science Foundation Grants 0456335 and 0336951. help reconcile the observed discrepancies. A hydrated transition This is contribution no. 07-04 of the Berkeley Seismological Laboratory, would help to lower the transition zone shear velocities, as required, University of California.

1. Dziewonski AM, Hager BH, O’Connell RJ (1977) J Geophys Res 82:239–255. 33. Laske G, Masters G (1998) Geophys J Int 133:508–520. 2. Woodhouse JH, Dziewonski AM (1984) J Geophys Res 89:9793–9823. 34. Liu W, Kung J, Li B (2005) Geophys Res Lett 32:L16301. 3. Ritsema J, van Heijst HJ, Woodhouse JH (2004) J Geophys Res 109(B0):2302. 35. Duffy TS, Zha C, Downs RT, Mao H, Hemley RJ (1995) Nature 378:170–173. 4. Grand SP (2002) Philos Trans R Soc London A 360:2475–11169. 36. Smyth JR, Jacobsen SD (2006) Geophys Monogr AM Geophys Union 168:1–12. 5. Panning M, Romanowicz B (2006) Geophys J Int 167:361–379. 37. Li B, Liebermann RC, Weidner DJ (1998) Science 281:675–677. 6. Li XD, Romanowicz B (1996) J Geophys Res 101:22245–22272. 38. Katsura T, Mayama N, Shouno K, Sakai M, Yoneda A, Suzuki I (2001) Phys Earth 7. Goes S, Govers R, Vacher P (2000) J Geophys Res 105:11153–11169. Planet Inter 124:163–166. 8. Goes S, van der Lee S (2002) J Geophys Res 107(B3):2050. 39. Weidner DJ, Wang Y (2000) Geophys Monogr Am Geophys Union 117:215–235. 9. Deschamps F, Trampert J, Snieder R (2002) Phys Earth Planet Inter 129:245–264. 40. Vacher P, Mocquet A, Sotin C (1998) Phys Earth Planet Inter 106:275–298. 10. Forte AM, Perry HKC (2000) Science 290:1940–1944. 41. Liebermann RC (2000) Geophys Monogr Am Geophys Union 117:181–199. 11. Trampert J, Deschamps F, Resovsky J, Yuen D (2004) Science 306:853–856. 42. Bass JD, Anderson DL (1984) Geophys Res Lett 11:237–240. 12. Forte AM, Mitrovica JX (2001) Nature 410:1049–1056. 43. Anderson DL, Bass JD (1986) Nature 320:321–328. 13. Ishii M, Tromp J (1999) Science 285:1231–1236. 44. Duffy TS, Anderson DL (1989) J Geophys Res 94:1895–1912. 14. Cammarano F, Goes S, Giardini D, Vacher P (2003) Phys Earth Planet Inter 45. Hofmann AW (2003) in Treatise on Geochemistry, eds Holland HD, Turekian KK 138:197–222. (Elsevier–Pergamon, Oxford), Vol 2, pp 61–101. 15. Karato SI (1993) Geophys Res Lett 20:1623–1626. 46. van Keken PE, Hauri EH, Ballentine CJ (2002) Annu Rev Earth Planet Sci 16. Stixrude L, Lithgow-Bertelloni C (2005) Geophys J Int 162:610–632. 30:493–525. 17. Fiquet G (2001) Z Kristallogr 216:248–271. 47. Bercovici D, Karato S (2003) Nature 425:39–44. 18. Jackson I (2000) Geophys Monogr Am Geophys Union 117:265–289. 48. Jaques A, Green D (1980) Contrib Mineral Petrol 73:287–310. 19. Dziewonski AM, Anderson DL (1981) Phys Earth Planet Inter 25:297–356. 49. Meibom A, Anderson DL (2003) Earth Planet Sci Lett 217:123–139. 20. Kennett BLN, Engdahl ER, Buland RP (1995) Geophys J Int 122:108–124. 50. Hirschmann MM, Stolper EM (1996) Contrib Mineral Petrol 124:185–208. 21. Cammarano F, Goes S, Deuss A, Giardini D (2005) Earth Planet Sci Lett 232:227–243. 51. Stixrude L, Lithgow-Bertelloni C (2005) J Geophys Res 110(B0):3204. 22. Cammarano F, Deuss A, Goes S, Giardini D (2005) J Geophys Res 110(B0):1306. 52. Tackley P, Xie S, Nakagawa T, Hernlund JW (2005) Geophys Monogr Am Geophys 23. Me´gnin C, Romanowicz B (2000) Geophys J Int 138:366–380. Union 160:83–99. 24. Gung F, Romanowicz B (2004) Geophys J Int 157:813–830. 53. Kohlstedt DL, Keppler H, Rubie DC (1996) Contrib Mineral Petrol 123:345–357. 25. Li XD, Romanowicz B (1995) Geophys J Int 121:695–709. 54. Smyth JR, Holl CM, Frost DC, Jacobsen SD, Langenhorst F, McCammon CA (2003) 26. Bassin C, Laske G, Masters G (2000) Trans Am Geophys Union 81:F897. Am Mineral 88:1402–1407. 27. Tarantola A, Valette B (1982) Rev Geophys Space Phys 20:219–232. 55. Ohtani E, Litasov K, Hosoya T, Kubo T, Kondo T (2004) Phys Earth Planet Inter 28. Christensen U (1984) Geophys J R Astron Soc 77:343–384. 143–144:255–269. 29. Goes S, Simons FJ, Yoshizawa K (2005) Earth Planet Sci Lett 236 227–237. 56. Karato SI (1995) Proc Jpn Acad 71:203–207. 30. Turcotte DL, Schubert G (1982) Geodynamics (Cambridge Univ Press, Cambridge, UK). 57. Jacobsen SD, Smyth JR, Spetzler H, Holl CM, Frost DJ (2004) Phys Earth Planet Inter 31. Chapman DS (1986) Geol Soc Spec Publ 24:63–70. 143–144:47–56. 32. Marone F, Romanowicz B (2006) Geophys J Int, in press. 58. Lawrence JF, Shearer PM (2006) Geochem Geophys Geosyst 7:Q10012.

9144 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.0608075104 Cammarano and Romanowicz Downloaded by guest on September 27, 2021